SK Hynix Plans On Integrating GPU & Reminiscence Semiconductors Right into a Single Package deal, Defying Trade Tendencies

The Korean reminiscence producer SK Hynix has determined to interrupt from the business’s custom of semiconductor growth, as they plan to combine reminiscence and logic semiconductors on the identical die.
SK Hynix Believes That Interconnectivity Via Packaging Strategies Is “Inefficient,” Plans On Taking A Completely different Route
In line with stories by Korean shops, SK Hynix believes there’s a risk of a extra “environment friendly” implementation of semiconductor placements out there, which the agency believes they will obtain with next-gen HBM4 reminiscence. It’s stated that SK Hynix has employed a substantial variety of design consultants for logic semiconductors, which within the first place did not make sense for the reason that firm solely focuses on reminiscence, however with right this moment’s growth, it looks like SK Hynix needs it to tackle the semiconductor markets. What has prompted SK Hynix for a “speedy” change within the ecosystem? Properly, we’ll talk about this afterward.
Concerning current-gen merchandise, superior reminiscence semiconductors equivalent to HBM are attaining effectivity by attaching them as carefully as potential subsequent to GPU chips, that are logic semiconductors. Particular person computations are separated via a devoted chip, which, in broader phrases, does seem like an inefficient methodology. Packaging methods such because the CoWoS turn out to be useful to bridge the hole between the semiconductors. Nevertheless, some form of “disparity” remains to be current, which is now meant to be catered by built-in reminiscence and logic semiconductors right into a single piece.
Shifting on to why SK Hynix wishes such implementation, it’s easy. Presently, the semiconductor ecosystem order is split into chip design (fabless), consignment manufacturing (foundry), and reminiscence/logic. This not solely includes subtle gear for every manufacturing course of, however typically, the work needs to be outsourced to completely different corporations specializing of their phases of growth. A significant downside behind that is the “elevated” reliance on particular person corporations, which has been evident in fashionable instances when it comes to large order backlogs, which slows down the general retailing course of. Since corporations like NVIDIA intend to tempo up the manufacturing to-delivery course of, lowering dependencies on one another is a main consider attaining the top purpose.

It hasn’t been revealed how SK Hynix plans on integrating logic and reminiscence semiconductors right into a single package deal, however sources say that SK Hynix is understood to be discussing the HBM4 design methodology with a number of international fabless corporations, together with NVIDIA. We might see some form of a “joint design” from the concerned corporations, which might act as a breakthrough when it comes to computational efficiency in addition to dashing up the general manufacturing course of, which is a crucial want for contemporary instances.


Primarily based on historical past, course of shrinking was seen as the one option to obtain generational beneficial properties, as validated by Moore’s Regulation, nevertheless it seems like the longer term is not wholly depending on it. It’s stated that the semiconductor/reminiscence business might see a “megaton-level wave” that would cross all boundaries outlined by current-gen limitations if corporations can pull off the above-discussed concept. The long run holds loads for everybody, and will probably be thrilling to see how corporations pave the best way in the direction of next-gen efficiency.
Information Supply: Joongang